GDPx benchmark leaves the best AI model at just 12% - Tustin AI Daily Briefing, Nov 2

Tustin AI Daily Briefing · Nov 2, 2026 · 6:31

Listen in the Podli app 🎧

Follow your favourite podcasts, listen offline and in the car with CarPlay and Android Auto, and always pick up where you left off. Free to try.

Stories in today's episode — tap a headline to read the full piece:

SemiAnalysis launches GDPx, a frontier agentic benchmark where the best model scores just 12%
OpenAI raises about $11 billion of a $30 billion target and cuts valuation to $1.25 trillion
Sierra added over $150M of ARR in one quarter while ServiceNow fell 26.9% in a day
The datacenter backlash has gone bipartisan ahead of the Nov. 3 midterms
Blackstone tells investors Anthropic may get a "junk" credit rating after its IPO
YC's Winter 2027 Request for Startups: eleven categories, zero consumer, all about AI's bottlenecks
Rewriting one title and subtitle took agent citations from 2 of 240 sessions to 240 of 240
Binance hack confirmed at $6.36B, attributed to North Korea — and issuers froze about 78% of the stablecoins
GTM software is the B2B category public markets hate most
Realtime-Bench: frontier models top out at 36% on real-time interactive tasks
DEA grants federal registration to more than 900 medical cannabis dispensaries — but excludes dual-use states
The AI "control plane" war: eight tests to run before you buy an enterprise agent
Five controls to set before an AI agent can buy, pay, or cancel for you
Casado: AI just made kernel-level vulnerability hunting "basically free" again
Garry Tan: Particle-1 matches much larger models on agentic retrieval at 1/23 the price
Mollick: a GDPval successor needs human baselines — and GPT-6 Pro caught its grading errors
Allie K. Miller: her agent logged into a vendor system, verified every video, and drafted the reply — unprompted
Nate B Jones: expect a post-trained "attack model" by Christmas
PitchBook Direct Data: private-market feeds positioned as LLM grounding infrastructure
Karpathy ships on Opus 5.5 "medium" — and the effort-level debate becomes tabs-vs-spaces
The Information: enterprises resist bigger AI commitments as switching gets easier
Lenny's Newsletter: how Figma's PMs built PM OS, a shared Claude Code repo
OpenAI's Grid Protectors Program: compute campuses that cut demand when the grid is strained
Andrej Karpathy: HackerTube, built with Claude in about two hours